Predicting progression from depression to bipolar disorder and schizophrenia spectrum disorders: a comparison of prediction models with and without polygenic scores

Individuals with bipolar (BP) or schizophrenia spectrum disorders (SZspec) frequently receive a diagnosis of depression prior to their first BP or SZspec episode. However, depression is common and identifying which patients with depression are in the early stages of a BP or SZspec illness is challenging. This issue is clinically relevant, as early identification of BP and SZspec can lead to better treatment and improved patient outcomes. Our aim was to determine whether polygenic scores (PGS) can help identify which patients with depression will go on to develop BP or SZspec by comparing prediction models with and without PGS. Data were obtained from the iPSYCH2015 sample, which includes all individuals born in Denmark between 1981-2008 who were diagnosed with mood and SZspec disorders in psychiatric hospitals between 1995-2015. Our sample included 29,361 individuals diagnosed with depression (ICD-10 codes F32, F33) in inpatient, outpatient or emergency care, who had not previously been diagnosed with BP (F30-F31) or SZspec (F20-F29). PGS for major depression (PGS-MD), BP (PGS-BP), and schizophrenia (PGS-SZ) were calculated using a multi-PRS method. Prediction models with and without PGS were built using flexible parametric proportional hazards models with one knot combined with 10-fold cross validation. Predictors besides PGS included family history of depression, BP and SZ; clinical characteristics of the first depression episode (severity, psychotic symptoms, age at diagnosis, treatment setting, and treatment duration); prior or comorbid anxiety, alcohol or drug abuse diagnoses; prior prescriptions of antidepressant or antipsychotic medication; and demographic/socioeconomic characteristics (sex, urbanicity, occupation, education, parental income, and marital status). Patients were followed from 8 weeks after the end date of their first depression episode until the start date of a contact for BP or SZspec, death, emigration or December 31, 2018, whichever came first. Length of follow-up ranged from 1 day to 23 years. Of the 29,361 patients with depression, 975 subsequently received a BP diagnosis and 2,228 received a SZspec diagnosis. The strongest predictors of progression to BP were family history of BP [Hazard Ratio=2.79, 95% Confidence Interval=2.17-3.59], prior antidepressant use [1.43, 1.24-1.65], emergency care [1.40, 1.21-1.61], and female sex [1.35, 1.16-1.57]. The strongest predictors of progression to SZspec were age at depression diagnosis [HR= 0.90, 95% CI=0.89-0.91], psychotic symptoms [2.32, 1.92-2.80], male sex [1.40, 1.28-1.53], and prior antipsychotics use [1.55, 1.37-1.75]. PGS-BP and PGS-SZ were associated with progression to BP [PGS-BP: 1.17, 1.08-1.27; PGS-SZ: 1.13, 1.05-1.23], while PGS-SZ and PGS-MDD were associated with progression to SZspec [PGS-SZ: 1.17, 1.11-1.23; PGS-MD: 1.06, 1.01-1.11]. The models without PGS explained 14.4% and 19.9% of the variance (Royston's measure of explained variation) in progression to BP and SZspec, respectively. Adding PGS to the models improved the model performance by 14.4% for progression to BP, and by 5.4 % for progression to SZspec. These results suggest that PGS could potentially aid in predicting which patients with depression will progress to bipolar disorder and, to a lesser extent, schizophrenia spectrum disorders. Further research is needed to establish whether these effects are clinically meaningful.